AI Augmentation

Use this workflow when adding AI features to an existing product, evaluating integration readiness, assessing data availability, or planning LLM architecture.

Pre-Investigation

1. Define the specific AI use case.
2. Understand what data exists and whether it is accessible.
3. Confirm latency, budget, and privacy constraints.
4. Check for existing ML or AI pipelines.

Investigation Phases

Phase	Focus	Evidence
Data assessment	Inventory, storage, access patterns, quality, sensitivity.	Volumes, freshness, sample records, permissions.
Integration architecture	API endpoints, async jobs, realtime needs, auth.	Candidate flows and current infrastructure constraints.
Implementation path	Direct calls, cached embeddings, fine-tuning, RAG, evaluation.	Cost estimate, latency measurements, fallback strategy.

Checklist

• Complete data inventory.
• Assess data quality.
• Review privacy and compliance constraints.
• Evaluate provider options.
• Identify integration points.
• Estimate token, storage, and compute costs.
• Define fallback behavior.
• Create an evaluation plan.

Common Risks

• Data is not AI-ready.
• Latency conflicts with product expectations.
• Token usage grows unexpectedly.
• Model hallucinations affect user outcomes.
• Provider lock-in is introduced too early.